fix(responses): complete done-terminated streams without finish_reason

Allow response.completed to be emitted on [DONE] when the stream already produced valid Responses output, so clean EOF does not trigger false retry loops.

Author: NGLSL

internal/translator/openai/openai/responses/openai_openai-responses_response.go

@@ -62,6 +62,19 @@ func emitRespEvent(event string, payload []byte) []byte {
 	return translatorcommon.SSEEventData(event, payload)
 }
 
+func hasCompletedResponseOutput(st *oaiToResponsesState) bool {
+	if st == nil {
+		return false
+	}
+	if len(st.MsgItemAdded) > 0 || len(st.FuncCallIDs) > 0 || len(st.Reasonings) > 0 {
+		return true
+	}
+	if st.ReasoningID != "" && st.ReasoningBuf.Len() > 0 {
+		return true
+	}
+	return false
+}
+
 func buildResponsesCompletedEvent(st *oaiToResponsesState, requestRawJSON []byte, nextSeq func() int) []byte {
 	completed := []byte(`{"type":"response.completed","sequence_number":0,"response":{"id":"","object":"response","created_at":0,"status":"completed","background":false,"error":null}}`)
 	completed, _ = sjson.SetBytes(completed, "sequence_number", nextSeq())
@@ -146,6 +159,12 @@ func buildResponsesCompletedEvent(st *oaiToResponsesState, requestRawJSON []byte
 			outputItems = append(outputItems, completedOutputItem{index: r.OutputIndex, raw: item})
 		}
 	}
+	if st.ReasoningID != "" {
+		item := []byte(`{"id":"","type":"reasoning","summary":[{"type":"summary_text","text":""}]}`)
+		item, _ = sjson.SetBytes(item, "id", st.ReasoningID)
+		item, _ = sjson.SetBytes(item, "summary.0.text", st.ReasoningBuf.String())
+		outputItems = append(outputItems, completedOutputItem{index: st.ReasoningIndex, raw: item})
+	}
 	if len(st.MsgItemAdded) > 0 {
 		for i := range st.MsgItemAdded {
 			txt := ""
@@ -227,7 +246,7 @@ func ConvertOpenAIChatCompletionsResponseToOpenAIResponses(ctx context.Context,
 		return [][]byte{}
 	}
 	if bytes.Equal(rawJSON, []byte("[DONE]")) {
-		if st.CompletionPending && !st.CompletedEmitted {
+		if !st.CompletedEmitted && (st.CompletionPending || hasCompletedResponseOutput(st)) {
 			st.CompletedEmitted = true
 			return [][]byte{buildResponsesCompletedEvent(st, requestRawJSON, func() int { st.Seq++; return st.Seq })}
 		}

internal/translator/openai/openai/responses/openai_openai-responses_response_test.go

@@ -30,13 +30,17 @@ func TestConvertOpenAIChatCompletionsResponseToOpenAIResponses_ResponseCompleted
 	request := []byte(`{"model":"gpt-5.4","tool_choice":"auto","parallel_tool_calls":true}`)
 
 	tests := []struct {
-		name           string
-		in             []string
-		doneInputIndex int // Index in tt.in where the terminal [DONE] chunk arrives and response.completed must be emitted.
-		hasUsage       bool
-		inputTokens    int64
-		outputTokens   int64
-		totalTokens    int64
+		name                  string
+		in                    []string
+		doneInputIndex        int // Index in tt.in where the terminal [DONE] chunk arrives and response.completed must be emitted.
+		hasUsage              bool
+		inputTokens           int64
+		outputTokens          int64
+		totalTokens           int64
+		wantMessageText       string
+		wantFunctionCallID    string
+		wantFunctionCallName  string
+		wantFunctionArguments string
 	}{
 		{
 			// A provider may send finish_reason first and only attach usage in a later chunk (e.g. Vertex AI),
@@ -138,6 +142,85 @@ func TestConvertOpenAIChatCompletionsResponseToOpenAIResponses_ResponseCompleted
 	}
 }
 
+func TestConvertOpenAIChatCompletionsResponseToOpenAIResponses_MessageWithoutFinishReasonCompletesOnDone(t *testing.T) {
+	t.Parallel()
+
+	request := []byte(`{"model":"gpt-5.4","tool_choice":"auto","parallel_tool_calls":true}`)
+	in := []string{
+		`data: {"id":"resp_msg_no_finish","object":"chat.completion.chunk","created":1773896263,"model":"model","choices":[{"index":0,"delta":{"role":"assistant","content":"hello world","reasoning_content":null,"tool_calls":null},"finish_reason":null}]}`,
+		`data: [DONE]`,
+	}
+
+	var param any
+	completedInputIndex := -1
+	var completedData gjson.Result
+	for i, line := range in {
+		for _, chunk := range ConvertOpenAIChatCompletionsResponseToOpenAIResponses(context.Background(), "model", request, request, []byte(line), &param) {
+			event, data := parseOpenAIResponsesSSEEvent(t, chunk)
+			if event != "response.completed" {
+				continue
+			}
+			completedInputIndex = i
+			completedData = data
+		}
+	}
+
+	if completedInputIndex != 1 {
+		t.Fatalf("expected response.completed on terminal [DONE] chunk at input index 1, got %d", completedInputIndex)
+	}
+	if got := completedData.Get("response.output.0.type").String(); got != "message" {
+		t.Fatalf("unexpected response.output.0.type: got %q want %q", got, "message")
+	}
+	if got := completedData.Get("response.output.0.content.0.text").String(); got != "hello world" {
+		t.Fatalf("unexpected response.output.0.content.0.text: got %q want %q", got, "hello world")
+	}
+}
+
+func TestConvertOpenAIChatCompletionsResponseToOpenAIResponses_FunctionCallWithoutFinishReasonCompletesOnDone(t *testing.T) {
+	t.Parallel()
+
+	request := []byte(`{"model":"gpt-5.4","tool_choice":"auto","parallel_tool_calls":true}`)
+	in := []string{
+		`data: {"id":"resp_func_no_finish","object":"chat.completion.chunk","created":1773896263,"model":"model","choices":[{"index":0,"delta":{"role":"assistant","content":null,"reasoning_content":null,"tool_calls":[{"index":0,"id":"call_func_no_finish","type":"function","function":{"name":"read","arguments":""}}]},"finish_reason":null}]}`,
+		`data: {"id":"resp_func_no_finish","object":"chat.completion.chunk","created":1773896263,"model":"model","choices":[{"index":0,"delta":{"role":null,"content":null,"reasoning_content":null,"tool_calls":[{"index":0,"function":{"arguments":"{\"filePath\":\"C:\\\\repo\\\\README.md\"}"}}]},"finish_reason":null}]}`,
+		`data: [DONE]`,
+	}
+
+	var param any
+	completedInputIndex := -1
+	var completedData gjson.Result
+	for i, line := range in {
+		for _, chunk := range ConvertOpenAIChatCompletionsResponseToOpenAIResponses(context.Background(), "model", request, request, []byte(line), &param) {
+			event, data := parseOpenAIResponsesSSEEvent(t, chunk)
+			if event != "response.completed" {
+				continue
+			}
+			completedInputIndex = i
+			completedData = data
+		}
+	}
+
+	if completedInputIndex != 2 {
+		t.Fatalf("expected response.completed on terminal [DONE] chunk at input index 2, got %d", completedInputIndex)
+	}
+	if got := completedData.Get("response.output.0.type").String(); got != "function_call" {
+		t.Fatalf("unexpected response.output.0.type: got %q want %q", got, "function_call")
+	}
+	if got := completedData.Get("response.output.0.call_id").String(); got != "call_func_no_finish" {
+		t.Fatalf("unexpected response.output.0.call_id: got %q want %q", got, "call_func_no_finish")
+	}
+	if got := completedData.Get("response.output.0.name").String(); got != "read" {
+		t.Fatalf("unexpected response.output.0.name: got %q want %q", got, "read")
+	}
+	args := completedData.Get("response.output.0.arguments").String()
+	if !gjson.Valid(args) {
+		t.Fatalf("expected response.output.0.arguments to be valid JSON, got %q", args)
+	}
+	if got := gjson.Get(args, "filePath").String(); got != `C:\repo\README.md` {
+		t.Fatalf("unexpected response.output.0.arguments.filePath: got %q want %q", got, `C:\repo\README.md`)
+	}
+}
+
 func TestConvertOpenAIChatCompletionsResponseToOpenAIResponses_MultipleToolCallsRemainSeparate(t *testing.T) {
 	in := []string{
 		`data: {"id":"resp_test","object":"chat.completion.chunk","created":1773896263,"model":"model","choices":[{"index":0,"delta":{"role":"assistant","content":null,"reasoning_content":null,"tool_calls":[{"index":0,"id":"call_read","type":"function","function":{"name":"read","arguments":""}}]},"finish_reason":null}]}`,